Scholarlink

<p>Accurate estimation of energy requirements for biomass pyrolysis is essential for designing cost‑efficient and sustainable thermochemical conversion systems. This study addresses the challenge of predicting pyrolysis energy requirement by integrating comprehensive feedstock compositional data with process operational parameters, analyzed through a machine learning (ML)‑based framework. A curated dataset of 633 experimentally validated records from peer‑reviewed publications was compiled, encompassing elemental composition (C, H, N, S, O, and ash content), biochemical composition (protein, lipid, and carbohydrate), and operational parameters. Models were trained and validated using a 9:1 split with five-fold cross‑validation to ensure robust generalization. Eight algorithms, including decision tree, adaptive boosting (AdaBoost), random forest, <i>K</i>‑nearest neighbors (KNN), ensemble learning, convolutional neural network (CNN), support vector regression (SVR), and multilayer perceptron (MLP), were optimized via hyperparameter tuning and evaluated through the coefficient of determination (<i>R</i><sup>2</sup>), mean squared error (<i>E</i><sub>MS</sub>), and average absolute relative error (<i>E</i><sub>AAR</sub>). Results demonstrated that AdaBoost and random forest achieved superior generalization on unseen data (test <i>R</i><sup>2</sup>≥0.893 and test <i>E</i><sub>AAR</sub>≤7.18%), with AdaBoost outperforming all standalone approaches by achieving the lowest test error rates. SHapley Additive exPlanations (SHAP) analysis identified ash content, temperature, and elemental composition (S, O, and C) as the most influential features driving model predictions. The analysis indicated a strong negative association between high ash content and the predicted energy requirement, while higher temperatures and combustible elements exhibited a positive correlation with the model’s output. The proposed data‑driven approach offers a reliable predictive tool for energy requirement estimation, supporting informed process design, feedstock selection, and optimization strategies for industrial biomass pyrolysis.</p> Graphical abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine-learning-based prediction of energy requirements for biomass pyrolysis via compositional and operational parameters

  • 期刊论文

  • 发表日期：

    2026年4月24日

查看全文
  • Anber Abraheem Shlash Mohammad,
  • Suleiman Ibrahim Mohammad,
  • Asokan Vasudevan,
  • Shaker Mohammed,
  • H. Malathi,
  • Rajashree Panigrahi,
  • H. Jemmy Christy,
  • Vimal Arora,
  • Debasish Shit,
  • Samim Sherzod

摘要

Accurate estimation of energy requirements for biomass pyrolysis is essential for designing cost‑efficient and sustainable thermochemical conversion systems. This study addresses the challenge of predicting pyrolysis energy requirement by integrating comprehensive feedstock compositional data with process operational parameters, analyzed through a machine learning (ML)‑based framework. A curated dataset of 633 experimentally validated records from peer‑reviewed publications was compiled, encompassing elemental composition (C, H, N, S, O, and ash content), biochemical composition (protein, lipid, and carbohydrate), and operational parameters. Models were trained and validated using a 9:1 split with five-fold cross‑validation to ensure robust generalization. Eight algorithms, including decision tree, adaptive boosting (AdaBoost), random forest, K‑nearest neighbors (KNN), ensemble learning, convolutional neural network (CNN), support vector regression (SVR), and multilayer perceptron (MLP), were optimized via hyperparameter tuning and evaluated through the coefficient of determination (R2), mean squared error (EMS), and average absolute relative error (EAAR). Results demonstrated that AdaBoost and random forest achieved superior generalization on unseen data (test R2≥0.893 and test EAAR≤7.18%), with AdaBoost outperforming all standalone approaches by achieving the lowest test error rates. SHapley Additive exPlanations (SHAP) analysis identified ash content, temperature, and elemental composition (S, O, and C) as the most influential features driving model predictions. The analysis indicated a strong negative association between high ash content and the predicted energy requirement, while higher temperatures and combustible elements exhibited a positive correlation with the model’s output. The proposed data‑driven approach offers a reliable predictive tool for energy requirement estimation, supporting informed process design, feedstock selection, and optimization strategies for industrial biomass pyrolysis.

Graphical abstract